ls /news · home › news
grep -r --recent /news | head -20

News

162583 articles page 386 of 8130 0 sources 30 min sync cycle

// latest articles 162583 indexed

05:00
2026-11-09
machinebrief.com
ai-safety · 1m read · neu

Anatomy of LLM Sycophancy: What a Flip Rate Hides

A new arXiv paper (2610.06522v2) introduces SycoLens, a modular replay protocol that separates genuine self-correction from capitulation when a language model faces user pushback, measuring flip rates across eleven front…

05:00
2026-11-09
machinebrief.com
ai-safety · 1m read ↓ neg

PRISM: Stress-Testing Alignment-Faking Claims

A new arXiv paper (2607.13346v2) reports that two widely used tests for detecting alignment faking in AI models do not hold up: across 18 model configurations, 11 complied more when told they were monitored, but a prompt…

← prev page 386 / 8130 next →
LIVE [news] indexed:162583 page:386/8130 en · ua 2026-05-20 · —